基于AI机器视觉技术的新能源无人值守场站自动巡检方法  

Automatic Inspection Method of New Energy Unattended Yard Station Based on AI Machine Vision Technology

作　　者：曹瑞安 CAO Ruian(Shanghai Shenergy New Energy Investment Co.,Ltd.,Shanghai 200020,China)

机构地区：[1]上海申能新能源投资有限公司,上海200020

出　　处：《电力大数据》2024年第11期48-56,共9页Power Systems and Big Data

摘　　要：新能源无人值守场站巡检数据构成复杂,导致专业人员无法准确快速地判断巡检结果,为此,提出基于AI机器视觉技术的新能源无人值守场站自动巡检方法。该方法利用多个基本分类器组成的初级学习器对视觉图像类别进行初次判断,具体的集成学习采用随机森林方法,构建多个决策树分类器,基于输出结果中最频繁出现的标签,判断新能源无人值守场站图像所属类别;基于信息增益选择最优新能源无人值守场站视觉图像的划分特征,利用ID3决策树算法输出具体的巡检结果。在测试结果中,mAP始终稳定在0.95以上,对巡检图像进行分类的准确率在90%~98%之间,对异常图像识别耗时可保持在15 s以下,具有较高的可靠性。The composition of inspection data for new energy unmanned stations is complex,which makes it difficult for professionals to accurately and quickly judge the inspection results.Therefore,an automatic inspection method for new energy unmanned stations based on AI machine vision technology is proposed.This method uses a primary learner composed of multiple basic classifiers to make initial judgments on the visual image category.The specific ensemble learning adopts the random forest method to construct multiple decision tree classifiers.Based on the most frequently appearing label in the output results,it determines the category of the new energy unmanned station image;Based on information gain,select the optimal visual image segmentation features for unmanned new energy stations,and use the ID3 decision tree algorithm to output specific inspection results.In the test results,mAP remained stable above 0.95,with an accuracy rate of 90%to 98%for classifying inspection images.The recognition time for abnormal images can be kept below 15 seconds,demonstrating high reliability.

关 键 词：AI机器视觉技术 新能源无人值守场站 自动巡检 随机森林 信息增益 ID3决策树算法 

分 类 号：TM63[电气工程—电力系统及自动化]